Neural networks in the detection of early acute myocardial ischemia from canine echocardiographic radio frequency data

A backpropagation neural network is used to differentiate ultrasound radio-frequency data from five closed-chest mongrel dogs. Echocardiograms captured the canine myocardial tissue at six time stages starting at the onset of ischemia which was induced by circumflex coronary artery occlusion. The backpropagation network is trained using data from the ischemic region and from a nonischemic control region, as indicated by a cardiologist. The performance of the backpropagation network in discriminating ischemic from normal radio frequency data is shown to be reasonably good considering the size of the training data available. Classification performance should greatly increase with an increase of the number of examples